BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Do ImageNet Classifiers Generalize to ImageNet?
DTSTART:20200224T101500
DTEND:20200224T111500
DTSTAMP:20260406T224535Z
UID:75536b66ada7ad0432fcd389b757692cfde881ebb1345109fefc074a
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Ludwig Schmidt - UC Berkeley\nIC Faculty candidate\n\nAbst
 ract:\nProgress on the ImageNet dataset seeded much of the excitement arou
 nd the machine learning revolution of the past decade. In this talk\, we a
 nalyze this progress in order to understand the obstacles blocking the pat
 h towards safe\, dependable\, and secure machine learning.\n\nFirst\, we w
 ill investigate the nature and extent of overfitting on ML benchmarks thro
 ugh reproducibility experiments for ImageNet and other key datasets. Our r
 esults show that overfitting through test set re-use is surprisingly absen
 t\, but distribution shift poses a major open problem for reliable ML.\n\n
 In the second part\, we will focus on a particular robustness issue\, know
 n as adversarial examples\, and develop methods inspired by optimization a
 nd generalization theory to address this issue. We conclude with a large e
 xperimental study of current robustness interventions that summarizes the 
 main challenges going forward.\n\nBio\nLudwig Schmidt is a postdoctoral re
 searcher at UC Berkeley working with Moritz Hardt and Ben Recht. Ludwig’
 s research interests revolve around the empirical and theoretical foundati
 ons of machine learning\, often with a focus on making machine learning mo
 re reliable. Before Berkeley\, Ludwig completed his PhD at MIT under the s
 upervision of Piotr Indyk. Ludwig received a Google PhD fellowship\, a Mic
 rosoft Simons fellowship\, a best paper award at the International Confere
 nce on Machine Learning (ICML)\, and the Sprowls dissertation award from M
 IT.\n\nMore information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR